A Novel Precision Analysis Of Astigmatism: Using The Within-Subject Covariance Matrix To Calculate The Astigmatic Prediction Error Secondary To Biometer Imprecision.

Analysing the precision of orthogonal power components like KP0 and KP45 is statistically limited and not clinically meaningful; the repeatability of each is calculated independently, and the reader is left with two metrics that indirectly describe one clinical entity. Previous studies have attempted to construct repeatability ellipses using the co-ordinate differences, though this is statistically flawed. This study applies a novel bivariate repeatability analysis in a comparative study of the IOLMaster 700 and Eyestar 900.

Single surgeon ophthalmology clinic in Queensland, Australia

This was a randomised study of eyes undergoing routine cataract surgery. Subjects underwent two repeated measurements using both the IOLMaster 700 and Eyestar 900 in random order. Astigmatism was converted into orthogonal power components. Multivariate statistics, astigmatism theory, and precision theory were applied to develop a new repeatability analysis for astigmatism. The within-subject covariance matrix of component powers was scaled by a multivariate critical factor to calculate the 95% “test-true” deviation ellipse. Due to the origins of orthogonal powers, the Euclidean radii of this ellipse reflect the magnitude of an obliquely crossed “error” cylinder at the respective plane.

140 left eyes from 140 subjects satisfied inclusion. The average scaled radius (test-retest error magnitude) for the anterior cornea was 0.29 D for the Eyestar (max = 0.33 D @ 86°) and 0.38 D for the IOLMaster (max = 0.42 D @ 116°). These errors describe the potential astigmatic prediction error secondary to imprecision as an obliquely crossed cylinder at the anterior corneal plane. For the posterior cornea, the errors were 0.04 D (max 0.05 @ 57°) for the Eyestar and 0.12 D (max = 0.12 @ 122°).

The Eyestar 900 has superior anterior and posterior astigmatism precision. Even with the best biometers, imprecision alone can produce up to 0.42 D of anterior astigmatic prediction error. To the authors’ knowledge, this is the first truly bivariate analysis of keratometry precision.